Disputation Isabel Patino Mejía
am Donnerstag, 18. Oktober 2019, um 15:00 Uhr in Raum C215, Sand 14
Estimating Head Measurements from 3D Point Clouds
Berichterstatter 1: Prof. Dr. Andreas Zell
Berichterstatter 2: Prof. Dr.-Ing. Hendrik P. A. Lensch
Human head measurements are valuable in ergonomics, acoustics, medicine, computer vision, and computer graphics, among other fields. Such measurements are usually obtained using entirely or partially manual tasks, which is a cumbersome practice since the level of accuracy depends on the expertise of the person that takes the measurements. Moreover, manually acquired measurements contain less information from which new measurements can be deduced when the subject is no longer accessible. Therefore, in order to overcome these disadvantages, an approach to automatically estimate measurements from 3D point clouds, which are long-term representations of humans, has been developed and is described in the presented manuscript. The 3D point clouds were acquired using an RGBD sensor Asus Xtion Pro Live and KinFu (open-source implementation of KinectFusion). Qualitative and quantitative evaluations of the estimated measurements are presented. Furthermore, the feasibility of the developed approach was evaluated through a case study in which the estimated measurements were used to appraise the influence of anthropometric data on the computation of the interaural time difference.
Considering the promising results obtained from the estimation of measurements from 3D models acquired with the sensor Asus Xtion Pro Live and KinFu (plus the results reported in the literature) and the development of new RGBD sensors, a study of the influence of seven different RGBD sensors on the reconstruction obtained with KinFu is also presented. This study contains qualitative and quantitative evaluations of reconstructions of four diverse objects captured at different distances that range from 40 cm to 120 cm. Such range was established according to the operational range of the sensors. Furthermore, a collection of obtained reconstructions is available as a dataset in uni-tuebingen.de/en/138898.